GitHub Issues
Turn a GitHub issue into a working PR through your orchestrator.
The most common WTmag flow: you have a GitHub issue, and you want an agent to work on it in isolation and get to a PR.
Through the orchestrator
Tell your orchestrator what you want to work on. You can reference the issue by number, describe it, or ask it to pull open issues for review:
Fix the login bug in issue 456
The orchestrator understands the intent and runs:
wtmag create --github 456 -t issueWTmag fetches the issue with gh, makes a worktree, writes the brief, and starts the agent. The orchestrator tells you the worker ID and how to attach.
Directly
You can also skip the orchestrator and run the command yourself:
wtmag create --github 456 -t issueSame result. The orchestrator just saves you the context switch.
What the worker does
The default issue prompt tells the agent to:
- Read the issue and understand it
- Implement the change
- Commit
- Push the branch
- Open a PR with
gh - Report the PR URL back
The worker runs in the background. You can attach to it at any point to watch, intervene, or take over.
Adding your own instructions
You can append extra instructions with --prompt:
wtmag create --github 456 -t issue --prompt "Start by writing regression tests."A word of caution: adding --prompt replaces the built-in delivery instructions (implement, commit, push, PR). Agents usually drop steps if you don't include them. Only add --prompt when you want something specific on top — and include the full workflow in your text if you still want a PR at the end.
Don't use --prompt to restate the issue. WTmag already includes the issue text in the brief.
Picking an agent
WTmag uses the configured default. Override for one run:
wtmag create --github 456 -t issue --agent opencodeOr change the default in your config. See agents for the full list.
Picking a model
Model overrides work with Pi and OpenCode right now:
wtmag create --github 456 -t issue --agent opencode --model openai/gpt-5.2:highOther agents will fail instead of guessing flags. Set the model in the agent's own config for those.